We investigated the horizontal resolution dependence of atmospheric radionuclide (Cs-137) simulations of the Fukushima nuclear accident on March 15, 2011. We used Eulerian and Lagrangian transport models with low- (15-km), medium- (3-km), and high- (500-m) resolutions; both models were driven by the same meteorological analysis that was prepared by our data assimilation system (NHM-LETKF) for each horizontal resolution. This preparation was necessary for the resolution-dependent investigation, excluding any interpolation or averaging of meteorological fields. In the results, the 15-km grid analysis could not reproduce Fukushima’s mountainous topography in detail, and consequently failed to depict a complex wind structure over mountains and valleys. In reality, the Cs-137 plume emitted from the Fukushima Daiichi Nuclear Power Plant (FDNPP) was mostly blocked by Mt. Azuma and other mountains along the Naka-dori valley after crossing over Abukuma Mountains on March 15, 2011. However, the 15-km grid simulations could not represent the blockage of the Cs-137 plume, which unnaturally spread through the Naka-dori valley. In contrast, the 3-km and 500-m grid simulations produced very similar Cs-137 concentrations and depositions, and successfully produced the plume blockage and deposition along the Naka-dori valley. In conclusion, low-resolution (15-km grid or greater) atmospheric models should be avoided for assessing the Fukushima nuclear accident when a regional analysis is needed. Meanwhile, it is reasonable to use 3-km grid models instead of 500-m grid models due to their similarities and the high computational burden of 500-m grid model simulations.